Soil & Tillage Research 91 (2006) 173–180

www.elsevier.com/locate/still

Quantification of soil shrinkage in 2D by digital image

processing of soil surface
X. Peng a,*, R. Horn a, S. Peth a, A. Smucker b
a
Institute of Plant Nutrition and Soil Science, CAU Kiel, Olshausenstr. 40, 24118 Kiel, Germany
b
Department of Crop and Soil Science, Michigan State University, East Lansing, MI 48824, USA
Received 13 July 2005; received in revised form 14 November 2005; accepted 16 December 2005

Abstract
Knowledge of soil shrinkage behavior is needed to improve the understanding and prediction of changes of unsaturated
hydraulic properties in non-rigid soils. The heterogeneity and interaction of horizontal and vertical soil shrinkages that produce soil
cracks and associated soil subsidence require additional quantification. Vertical shrinkage can be calculated easily by soil height
with vernier caliper. However, a quantitative and feasible measurement of horizontal shrinkage has not been developed yet because
of the complicated and irregular geometry of soil cracks. This paper introduces a new method to measure soil cracks non-
destructively and continuously by digital image analysis. Using Adobe Photoshop and Windows Scion 4.02 image processing, the
proposed procedure accurately identifies changes as small as 1.0 mm2 and shows differences even when areas of soil cracks were
increased by as little as 1%. Various geometry factor values indicated soil shrinkage in the two dimensions was anisotropic during
the whole drying. During initial dehydration from saturation, only subsidence shrinkage could be identified. With the further
dehydration, the soil cracks developed and increased in size. These results suggest that the heterogeneity of soil shrinkage in 2D
should be taken into account when modeling the total soil shrinkage behavior and associated unsaturated hydraulic properties.
# 2006 Elsevier B.V. All rights reserved.

Keywords: Soil shrinkage; Geometry factor; Image analysis; Soil cracks; Scion imaging

1. Introduction Vertical shrinkage results in soil subsidence while

horizontal shrinkage produces soil cracks. Vertical soil
When investigating soil shrinkage and preferential subsidence is easy to measure by changes of soil height.
flow in non-rigid soils, one of the challenges is the However, the measurement of soil cracks is more
quantification of the geometry of soil cracks. Soil difficult due to their complicated and irregular
cracking occurs when soils, especially swelling clayey geometries. The geometry of soil cracks including
soils, are subjected to dehydration. The cracks in the surface crack area and crack volume within the soil
soil matrix, ranging from millimeter to centimeter matrix plays an important role in water flow during and
widths, are generally visible unlike micropores and after soil volume changes. Pre-existing soil cracks
capillary pores (Hillel, 1998). induce surface water to reach more directly into the
In non-rigid soils, two shrinkage components of the groundwater level as preferential flow (Liu et al., 2003)
vertical and horizontal directions can be quantified. and potentially lead to groundwater contamination.
Meanwhile, the stress during the development of soil
cracks rearranges soil particles altering soil structure
* Corresponding author. Tel.: +49 431 8803191;
and its associated mechanical properties (Baumgartl
fax: +49 431 8802940. et al., 2000). Therefore, semi-quantitative evaluations
E-mail address: [email protected] (X. Peng). of soil cracks have been developed to estimate the

0167-1987/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.still.2005.12.012
174 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180

degree of soil cracking. Sand displacement (Dasog and further, water loss was only found from evaporation. At
Shashidhara, 1993) and flexible plastic rulers and latex the end of air-dryness, soil cores reached a quasi
rubber filling (Ringrose-Voase and Sanidad, 1996) has equilibrium (i.e. the water loss per day was smaller than
been used to identify crack volume. Although these 1.0 g) and then they were shifted to be oven-dried from
methods were inexpensive and special equipment was 40, 70 to 105 8C step by step. During the whole drying
not required, accuracy was not very satisfying. Garnier process, soil water content was recorded by electronic
et al. (1997) and Braudeau et al. (1999) developed laser balance with an accuracy of 0.01 g. The vertical and
analyses to monitor the soil height and diameter horizontal deformations of soil cores were measured
continuously and non-destructively, but some soil more than 20 times simultaneously as follows.
cracks within the soil matrix were possibly missed.
Image analysis of CT scans has also been utilized to 2.2. Soil vertical shrinkage
investigate changes in soil structure. Vogel (1997) and
Gantzer and Anderson (2002) depicted soil pore Soil height was determined by a vernier caliper with
structure in three dimensions with the X-ray scanning an accuracy of 0.05 mm at nine defined locations on the
tomographic technique. Velde (1999) applied image soil surface at the respective water content. The change
analyses of photographs to investigate network com- of soil height multiplied by initial soil core area
plexities of soil surface cracks. Image analysis could be represents the vertical shrinkage (soil subsidence).
used to identify contrasting intensities between the
darker soil pores and the lighter colored soil solid. 2.3. Soil horizontal shrinkage
Computer analyses of digital images (CADI) of soil
cracks are also non-destructive to soil structure. We The horizontal shrinkage was accounted for the
hypothesized that CADI applied to the surfaces of volume of the soil cracks. We measured the cross-
drying soils, will quantitatively evaluate horizontal sectional surface area of the soil cracks using image
shrinkage. Therefore, our objective was to propose a analysis and its corresponding crack volume using a
new method for repeated, non-destructive measurement separate paraffin filling.
of horizontal shrinkage during soil dehydration. In order
to control the experiment easily, this measurement was 2.3.1. Image analysis
carried out in laboratory. Soil surface showing the cracks, within the soil
volume and surrounding the soil core between the soil
2. Materials and methods and the cylinder wall, were photographed with a Sonny
digital cameral under macro model, positioned at a
2.1. Soils constant distance of 15 cm above the soil surface
(Fig. 1(a)). The greatest contrast of pixel intensity
One Histosol and three Gleysols classified according between soil cracks and the surface soil was accom-
to World Reference Base for Soil Resources (FAO) with plished using Adobe Photoshop 7.0. Photo size was
different textures and located under pastures in the fixed as same size as the outer wall of cylinder using the
southern part of the federal state Schleswig-Holstein, cutter tool, e.g. 10.3 cm. Shrinkage cracks in the soil
Germany, were investigated. The three Gleysols were surface were outlined using the ‘‘magic wand’’ tool,
derived from glacial and lacrustaine sediments with which separated cracks from surface soil by the bimodal
illite as the main clay mineral. histogram, which automatically contrasted cracks from
Undisturbed soil cores (10 cm in diameter and soil (Fig. 1(b)). The marked area represented a new
6.1 cm in height, three replicates) were taken in April layer overlaying the origin layer after running the ‘‘new
2004 by the stainless steel cylinder with the wall of filling layer’’ function. On the new layer, some pseudo-
0.15 cm width. The field water content of the cracks whose pixel intensities were similar to those of
investigated soils reaches 89.0–93.2% of saturation. real soil cracks could be erased with ‘‘eraser’’ tool.
Soil cores were put into a box for avoiding evaporation Although this crack-identification procedure depended
during transport. They were water saturated from the upon the human intervention, with a little practice, it
bottom by capillary rise in laboratory. The swelling part became very objective. Since pixel intensities of surface
above the cylinder developed during wetting was cut in soil clearly contrasted with those of the soil cracks the
order to keep the initial volume and thereafter air-dried subjective error became negligible. When the binary
continuously. At the beginning of air-dryness, water photo became similar to the shape of the real soil cracks,
could be drained from soil cores. When soils dehydrated Fig. 1(a) and (c), it was copied to a new file. The binary
 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180 175

Fig. 1. Graphical procedure of measuring the cross-sectional area of soil surface cracks with taking digital photo (a), marking soil surface cracks (b),
and converting binary image (c).

photo, Fig. 1(c), portrays that the white color becomes 2.3.2. Standardization of imaging
the soil cracks while the black color was regarded as the A series of known areas cut from black solid circle
soil and core background. were used to standardize the digital imaging and image
Each image photo was thereafter processed by the processing procedures (Fig. 2(a)–(c)). A small black
Scion imaging software. Scion (Image 4.02) is public solid circle and a big open circle were generated by the
domain software for Windows (Scioncorp, 2000) and computer (Fig. 2(a)). The open circle simulated the
can be used to measure the surface areas of objects in a outer wall of the soil cylinders, when fixing the image
digital format (Dowdy et al., 1998). This program is photo later. The black solid circle was cut into many
available for Windows operating systems and can be small pieces emulating soil cracks (Fig. 2(b)). These
downloaded at the following website: http:// pieces were positioned on the center of the open circle
www.scioncorp.com. It can also be downloaded for and photographed and image processed as outlined
Apple at: http://rsb.info.nih.gov/nih-image. above (Fig. 2(c)).
Before image processing, the bimodal image was These imaging procedures were verified by a
inverted, so that cracks became black and soil white. A sensitivity analysis (Fig. 3(a) and (b)). Using a series
threshold pixel value for the black filling of the soil of known areas from 1 to 10 mm2, we increased with
cracks was identified and the Scion program auto- each step by 1 mm2, which a relative increase is 10%.
matically filled the soil cracks. All pixel intensities Another series of known areas, from 101 to 110 mm2
equal to or greater than this threshold level were were photographed and imaged. In this measurement,
displayed as black soil cracks and all other pixels were each step increased 1% while the absolute increase was
displayed in white as background. Accordingly, the ‘‘set still 1 mm2.
scale’’ command of Scion 4.02 was adjusted to the
photo size. The area value of black color was shown 2.3.3. Estimation of the crack volume
after running ‘‘measure’’ command, and the white Quantitative imaging of soil cracks is somewhat
background pixels were ignored. limited by the intrinsic opacity of the depths of each soil

Fig. 2. Image analysis verification of a known area of black solid circle with an open circle as a reference size (a), small pieces cut from the solid
circle emulating soil cracks (b), and the pieces and the background being converted to binary image (c).
176 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180

Fig. 3. Sensitivity analysis of image procedure of 10% relative increase (a), and 1% relative increase (b) while their absolute actual area increase is
1 mm2.

crack within the soil matrix. This apparent limitation 2.4. Heterogeneity of soil shrinkage
was investigated by estimating total volume of surface
cracks by filling them with low melting point liquid During its deformation, the relation between soil
paraffin (Brasher et al., 1966). The paraffin volume that subsidence and soil cracks in soil shrinking hetero-
filled the macropore crack volumes was defined as the geneity was described by Bronswijk’s (1990) dimen-
volume of soil cracks. Therefore, additional 15 sionless geometry factor, rs, defined by:
undisturbed soil cores were analysed correspondingly Vi
to relate the area of soil surface cracks to their volume ln Vs
rs ¼ zi (3)
within the soil matrix using paraffin-filling method. ln zs
During the five different degrees of dryness with three
replicates, a regression relationship was obtained where Vs and zs are the initial volume and height of soil
between the area of soil surface cracks and the paraffin core at saturation. Vi and zi are the volume and the height
volume within the soil matrix at a given soil. Using the of soil core at the ith step, respectively. Vi is calculated
relationship we calculated the volume of soil cracks for by:
a given area of soil surface cracks. The relation between Vi ¼ Vs
Vci
Vvi (4)
soil crack area and soil crack volume was normalized by
soil crack area ratio (rA) and crack volume ratio (rV) where Vvi is the volume of soil subsidence at the ith
when taking into account the original size of soil step. There were five cases to describe the relationship
samples: between soil subsidence and soil cracks: (1) in the case
Ai of subsidence without cracking, rs = 1.0, (2) in the case
rA ¼ (1) of cracking without subsidence, rs ! 1, (3) in the case
As
of predominant subsidence, 1.0 < rs < 3.0, (4) in the
Vci case of predominant cracking, rs > 3.0, and (5) in the
rV ¼ (2) case of isotropic shrinkage, rs = 3.0.
Vs
where Ai and Vci are the soil surface crack area and the 2.5. Soil analysis
soil crack volume at ith dryness, respectively. As and Vs
are the soil surface area and soil volume at saturation, Saturated hydraulic conductivity (Ks) of undisturbed
respectively. soil cores was measured by the falling head method
 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180 177

(Kutı́lek and Nielsen, 1994). Disturbed soil samples Six different images, generated by the computer, having
were ground to pass 2 mm mesh for analysing soil known areas from 0.20 to 7.07 cm2, were used to verify
physical and chemical properties. Particle size distribu- measured values by image analysis. The absolute errors
tion was measured by the pipette method after removing of both measured areas and actual areas were 0.01–
organic materials. Soil organic carbon was determined 0.07 cm2 and the relative errors among them ranged
by dry combustion and cation exchange capacity (CEC) from 0.99% to 7.01%. A trend was found that the larger
by the ammonium acetate method. actual area had a larger absolute error but a smaller
relative error.
3. Results and discussion Sensitivity analysis was used to verify image
analysis. A 1:1 linear relationship between measured
3.1. Verifying the image analysis procedure and actual data was shown in Fig. 3(a) but not in
Fig. 3(b) although the absolute change at each step in
Fig. 1(a)–(c) shows the procedure of measuring soil the two figures were the same (i.e. 1 mm2). In Fig. 3(a),
surface crack areas with the aid of image analysis. In this 1 mm2 accounted for 10% or more of total area. In
procedure, images of soil cracks were processed Fig. 3(b), however, 1 mm2 alteration meant only 1% or
separately by Adobe Photoshop software before being less of change in area. As mentioned in Table 1, the
analyzed by the Scion image. During the image measured data was a little higher than the actual data. A
manipulation, the area designated as shrinkage cracks, little overestimation was ascribed to some unavoidable
could automatically be identified by overlaying the coarse-grained edge of the original image (Ogawa et al.,
processed image, which was the threshold for the binary 1999). The overestimation value increases when the
image. Although it may be subjected to visual error, the ratio of the perimeter to the area is higher, which was
discrepancies between the two layers could be easily confirmed by the smaller area with the higher
identified, minimizing mistaken identification of soil overestimation in Table 1. These results indicated that
cracks. This procedure was used to establish the best edge the image analysis could measure as small as 1.0 mm2
detection for identifying all soil cracks and it assisted in (Fig. 3(a)) and differentiated only 1% area change
establishing the binary image, which eliminated potential (Fig. 3(b)), but not determine as small as 1% of actual
threshold-induced biases and initiated thresholding pixel area. In this case, it should be pointed out that the
values that were generated automatically. This procedure computer-generated photo showed that simulated
was necessary to confirm that the binary image was cracks were black and the background was white.
exactly the same as the original soil image. Therefore, Naturally, the dark colored soil cracks have no such
this image analysis could avoid the potential errors pronounced difference of pixel intensity to the light
induced by manual thresholding, which were identified colored solid soil. Theoretically speaking, the method
by Ogawa et al. (1999) and Velde (1999). could distinguish such minor area if the color of soil
Due to the irregular shape of soil cracks the image- cracks differed clearly from that of soil solid. Sensitivity
processed values needed standardization. A standardi- analyses confirmed the proposed procedure not only
zation process of irregular areas simulated soil cracks in monitored the development of soil surface cracks
Fig. 2(a)–(c). We identified absolute and relative image continuously and non-destructively but also yielded a
processing errors. The differences between the actual high accuracy and sensitivity.
area and measured area of image are reported in Table 1.
3.2. Anisotropic shrinkage of investigated four
Table 1
soils in 2D
Comparisons of the actual area and measured area of digital camera
photos Table 2 shows some physical and chemical properties
Circle Actual Measured Absolute Relative
of the investigated soils. Eutric Histosol and Histic
diameter area area error error Gleysol contain over 20% organic C while Calcic Gleysol
(cm) (cm2) (cm2) (cm2) (%) and Dystric Gleysol have only less than 1.5% (Table 2).
0.5 0.20 0.21 0.01 7.01 The investigated soils range from soil texture. The
1.0 0.79 0.81 0.03 3.18 dominance of the Eutric Histosol is sand. Histic Gleysol
1.5 1.77 1.79 0.02 1.34 is classified as a clay loam. Calcic Gleysol contains
2.0 3.14 3.22 0.08 2.55 577 g kg
1 of silt, while in the Dystric Gleysol the clay
2.5 4.91 4.98 0.07 1.43
fraction dominates. Saturated hydraulic conductivity (Ks)
3.0 7.07 7.10 0.07 0.99
in the two organic matter enriched soils is much higher
178 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180

Table 2
Some physical and chemical soil properties of investigated soils
Soil type Depth (cm) Sand Silt Clay (g kg
1) Organic C pH (H2O) CECa (mmol kg
1) Ksb (cm day
1)
Eutric Histosol 30–40 703 182 115 222 4.14 176 25.3
Histic Gleysol 40–50 260 341 399 201 5.36 244 3.61
Calcic Gleysol 30–40 74 577 349 14 7.25 91 0.34
Dystric Gleysol 50–70 200 151 649 5 4.1 59 0.54
a
Cation exchange capacity.
b
Saturated hydraulic conductivity.

than that of the two inorganic Gleysols by one or two saturation to oven-dryness. The different slope values
orders of magnitude, respectively. between the four soils ranging from 0.636 to 0.966 were
Measurement of the area of soil surface cracks was not ascribed to the variety of soil texture and structure. But
sufficient for determining the horizontal shrinkage. how soil properties influence the geometry of soil cracks
Primarily not all cracks could be categorized as needs further investigation due to limited data in this
rectangular shapes within the soil matrix and soil crack study. Fig. 5 shows the various values of geometry factor
dimensions were not same at different soil depths. during dehydration. Of the all four soils, geometry factor
Vertical cross-sections through soil cracks were reported values started from 1.0 at the beginning of hydration from
to be an inverted triangle shape in the soil profile saturation, which meant there was only vertical
(Ringrose-Voase and Sanidad, 1996), or surface cracks shrinkage. With continuing dryness, the geometry factor
were larger than those below the surface. Therefore, increased quickly but less than 3.0, which indicated both
additional experiments were completed to correlate areas vertical and horizontal shrinkage occurred, but vertical
of surface cracks with the total volume of cracks shrinkage still dominated at this water range. When soil
throughout the soil matrix (Fig. 4). A significant linear was subjected to further dehydration, the geometry factor
relationship was obtained linking crack volume to value reached approximate 3.0 for the Histic Gleysol and
measured areas of surface cracks (P < 0.001). We the Calcic Gleysol manifesting an isotropic shrinkage. A
conclude that the geometry of soil cracks retain similar geometry factor exceeding 3.0 was however found for the
shaped isosceles triangles during soil shrinkage from Eutric Histosol and the Dystric Gleysol, where the

Fig. 4. Linear regression relationship between crack area ratio and crack volume ratio of four soils studied.
 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180 179

Fig. 5. Changes of geometry factor (squares) and void ratio (circles) during the whole dehydration of four soils studied.

horizontal shrinkage became predominant. The variety of Dystric Gleysol. Therefore, the anisotropy of soil
geometry factor was attributed to the changes of effective shrinkage in vertical and horizontal components
stress during dehydration. At the beginning of drying, depended not only on soil types but also on soil water
water was firstly drained from macropores by gravity and content even for same soil type. How soil properties
thus only vertical shrinkage occurred. If soil drying affect the geometry factor needs further investigation.
proceeded, water was lost from macropores shifting to
finer pores. Hydraulic stress thus increased. Horizontal 4. Conclusions
and vertical shrinkages both appeared while vertical
direction was predominant at the moment. If soil dried This paper introduced an inexpensive, non-destruc-
intensively, only very fine micropores still contained tive and quantitative measurement of soil shrinkage.
water. The menisci force in the micropores produced With this method the development of soil surface cracks
higher hydraulic stress than before. Soil particles and during dehydration can be monitored. Non-destructive
microaggregates could be rearranged and oriented into a measurements of surface cracks could be used to
new platy unit (Horn et al., 2003). Consequently, identify crack volume. Various geometry factor values,
horizontal hydraulic stress increased faster than the from saturation to oven dryness, exhibited an aniso-
vertical. The horizontal shrinkage became dominant tropic soil shrinkage. At least in these soils, and possibly
instead of vertical shrinkage while geometry factor all soils, horizontal shrinkage needs to be investigated
values were higher than 3. Fig. 5 also shows soil independently from vertical shrinkage. The lack of this
shrinkage curve depicted by void ratio and moisture ratio. study is limited in laboratory. Further investigations are
The geometry factor corresponding to shrinkage phase the extension of image analysis to field measurement
could be identified. Structural shrinkage phase, where the and the influence of soil properties on the geometry of
change of void ratio is less than that of moisture ratio at soil cracks.
the beginning of drying (Peng and Horn, 2005), was
approximately corresponded to the range of <3 geometry Acknowledgement
factor. The geometry factor of the other shrinkage phases
was near 3 for the Histic Gleysol and the Calcic Gleysol The first author is highly indebted to the German
and a little more than 3 for the Eutric Histosol and the Research Foundation (DFG) for its grant PE1404/1-1.
180 X. Peng et al. / Soil & Tillage Research 91 (2006) 173–180

References transient evaporation experiment. Soil Sci. Soc. Am. J. 61, 1555–
1563.
Baumgartl, T., Rostek, J., Horn, R., 2000. Internal and external Hillel, D., 1998. Environmental Soil Physics. Academic Press, Lon-
don, pp. 118–120.
stresses affecting the water retention curve. In: Horn, R., Van
den Akker, J.J.H., Arvidsson, J. (Eds.), Subsoil Compaction: Horn, R., Way, T., Rostek, J., 2003. Effect of repeated tractor wheeling
Distribution, Processes and Consequences. Catena Verlag, on stress/strain properties and consequences on physical proper-
Reiskirchen, pp. 3–12. ties in structural arable soils. Soil Till. Res. 73, 101–106.
Kutı́lek, M., Nielsen, D.R., 1994. Soil Hydrology. Catena Verlag,
Brasher, B.R., Franzmeier, D.P., Valassis, V., Davidson, S.E., 1966.
Use of Saran Resin to coat natural soil clods for bulk density and Cremlingen, Germany, pp. 98–101.
moisture retention measurements. Soil Sci. 101, 108. Liu, C.W., Cheng, S.W., Yu, W.S., Chen, S.K., 2003. Water infiltration
Braudeau, E., Costantini, J.M., Bellier, G., Colleuille, H., 1999. New rate in cracked paddy soil. Geoderma 117, 169–181.
Ogawa, S., Baveye, P., Boast, C.W., Parlange, J.Y., Steenhuis, T.,
device and method for soil shrinkage curve measurement and
characterization. Soil Sci. Soc. Am. J. 63, 525–535. 1999. Surface fractal characteristics of preferential flow patterns in
Bronswijk, J.J.B., 1990. Shrinkage geometry of a heavy clay soil at field soils: evaluation and effect of image processing. Geoderma
88, 109–136.
various stresses. Soil Sci. Soc. Am. J. 54, 1500–1502.
Dasog, G.S., Shashidhara, G.B., 1993. Dimension and volume of Peng, X., Horn, R., 2005. Modeling soil shrinkage curve across a wide
cracks in a Vertisol under different crop covers. Soil Sci. 156, 424– range of soil types. Soil Sci. Soc. Am. J. 69, 584–592.
428. Ringrose-Voase, A.J., Sanidad, W.B., 1996. A method for measuring
the development of surface cracks in soils: application to crack
Dowdy, R.H., Smucker, A.J.M., Dolan, M.S., Ferguson, J.C., 1998.
Automated image analyses for separating plant roots from soil development after lowland rice. Geoderma 71, 245–261.
debris elutriated from soil cores. Plant Soil 200, 91–94. Scioncorp, 2000. Scion Image Manual. www.scioncorp.com
Gantzer, C.T., Anderson, S.H., 2002. Computed tomographic mea- Velde, B., 1999. Structure of surface cracks in soils and muds.
Geoderma 93, 101–124.
surement of macroporosity in chisel-disk and no-tillage seedbeds.
Soil Till. Res. 64, 101–111. Vogel, H.J., 1997. Morphological determination of pore connectivity
Garnier, P., Rieu, M., Boivin, P., Vauclin, M., Baveye, P., 1997. as a function of pore size using serial sections. Eur. J. Soil Sci. 48,
365–377.
Determining the hydraulic properties of a swelling soil from a